From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (qmail 23639 invoked by alias); 5 Dec 2019 20:40:46 -0000 Mailing-List: contact libc-alpha-help@sourceware.org; run by ezmlm Precedence: bulk List-Id: List-Subscribe: List-Archive: List-Post: List-Help: , Sender: libc-alpha-owner@sourceware.org Received: (qmail 23531 invoked by uid 89); 5 Dec 2019 20:40:45 -0000 Authentication-Results: sourceware.org; auth=none X-Spam-SWARE-Status: No, score=-21.8 required=5.0 tests=AWL,BAYES_00,GIT_PATCH_0,GIT_PATCH_1,GIT_PATCH_2,GIT_PATCH_3,KAM_SHORT,RCVD_IN_DNSWL_NONE,SPF_PASS autolearn=ham version=3.3.1 spammy= X-HELO: mail-vk1-f194.google.com DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=linaro.org; s=google; h=to:references:from:autocrypt:subject:message-id:date:user-agent :mime-version:in-reply-to:content-language:content-transfer-encoding; bh=3XvXcWd70y/ive+3KZJDosu+7u/xo/mpKzFVRvr6Wvw=; b=KndfSL+FpdUF0royvP3dV7+gaqvooS6uccd9HzfWvo2tstPdhLrNHGiqWSHiWHUc5s R+ue4bLikq2n7m9dak6th2/fBEDbFrD2QSPeYz5KyYrLMUz8SOkleeorc0sHLbF2VmD8 K8rcVtw1smr+UhiggqiKHzKWtH6CSI7Yih4i5dXF9bmkNwkjsCu2rqebMbe68qGyKFQN 9Pc5h8YWWovb3+7mlJ4RJ9COJR9zEIYi9bUCykVTr4UEJvGP1YKnjbMnfJS8r3LRvfkG nmCGNF8lR6BkP50API7XD7O/1Lh6h0xbgIbbVywPFXixmEpIv5IjvMfWg0ytCpY1Mpqz G3qA== Return-Path: To: libc-alpha@sourceware.org References: <1575297977-2589-1-git-send-email-stli@linux.ibm.com> <1575297977-2589-8-git-send-email-stli@linux.ibm.com> From: Adhemerval Zanella Subject: Re: [PATCH 07/13] Use GCC builtins for nearbyint functions if desired. Message-ID: <251cf3fb-afbc-1fa7-6209-9c3c7de1c918@linaro.org> Date: Thu, 05 Dec 2019 20:40:00 -0000 User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:68.0) Gecko/20100101 Thunderbird/68.2.1 MIME-Version: 1.0 In-Reply-To: <1575297977-2589-8-git-send-email-stli@linux.ibm.com> Content-Type: text/plain; charset=utf-8 Content-Transfer-Encoding: 8bit X-SW-Source: 2019-12/txt/msg00197.txt.bz2 On 02/12/2019 11:46, Stefan Liebler wrote: > This patch is using the corresponding GCC builtin for nearbyintf, nearbyint, > nearbintl and nearbyintf128 if the USE_FUNCTION_BUILTIN macros are defined to one > in math-use-builtins.h. > > This is the case for s390 if build with at least --march=z196 --mzarch. > Otherwise the generic implementation is used. The code of the generic > implementation is not changed except changes in code style. LGTM with some changes below. Reviewed-by: Adhemerval Zanella > --- > sysdeps/generic/math-use-builtins.h | 29 ++++++++ > sysdeps/ieee754/dbl-64/s_nearbyint.c | 8 +++ > sysdeps/ieee754/float128/float128_private.h | 4 ++ > sysdeps/ieee754/flt-32/s_nearbyintf.c | 74 ++++++++++++-------- > sysdeps/ieee754/ldbl-128/s_nearbyintl.c | 76 ++++++++++++--------- > sysdeps/s390/fpu/math-use-builtins.h | 45 ++++++++++++ > 6 files changed, 175 insertions(+), 61 deletions(-) > create mode 100644 sysdeps/generic/math-use-builtins.h > create mode 100644 sysdeps/s390/fpu/math-use-builtins.h > > diff --git a/sysdeps/generic/math-use-builtins.h b/sysdeps/generic/math-use-builtins.h > new file mode 100644 > index 0000000000..e12490ed41 > --- /dev/null > +++ b/sysdeps/generic/math-use-builtins.h > @@ -0,0 +1,29 @@ > +/* Using math gcc builtins instead of generic implementation. Generic version. > + Copyright (C) 2019 Free Software Foundation, Inc. > + This file is part of the GNU C Library. > + > + The GNU C Library is free software; you can redistribute it and/or > + modify it under the terms of the GNU Lesser General Public > + License as published by the Free Software Foundation; either > + version 2.1 of the License, or (at your option) any later version. > + > + The GNU C Library is distributed in the hope that it will be useful, > + but WITHOUT ANY WARRANTY; without even the implied warranty of > + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU > + Lesser General Public License for more details. > + > + You should have received a copy of the GNU Lesser General Public > + License along with the GNU C Library; if not, see > + . */ > + > +#ifndef MATH_USE_BUILTINS_H > +#define MATH_USE_BUILTINS_H 1 > + > +/* Define these macros to 1 to use __builtin_xyz instead of the > + generic implementation. */ > +#define USE_NEARBYINT_BUILTIN 0 > +#define USE_NEARBYINTF_BUILTIN 0 > +#define USE_NEARBYINTL_BUILTIN 0 > +#define USE_NEARBYINTF128_BUILTIN 0 > + > +#endif /* math-use-builtins.h */ Ok. > diff --git a/sysdeps/ieee754/dbl-64/s_nearbyint.c b/sysdeps/ieee754/dbl-64/s_nearbyint.c > index c261885c5a..08a148e21e 100644 > --- a/sysdeps/ieee754/dbl-64/s_nearbyint.c > +++ b/sysdeps/ieee754/dbl-64/s_nearbyint.c > @@ -26,16 +26,23 @@ > #include > #include > #include > +#include > > +#if ! USE_NEARBYINT_BUILTIN > static const double > TWO52[2] = { > 4.50359962737049600000e+15, /* 0x43300000, 0x00000000 */ > -4.50359962737049600000e+15, /* 0xC3300000, 0x00000000 */ > }; > +#endif You could just move it inside the !USE_NEARBYINT_BUILTIN within the function. > > double > __nearbyint (double x) > { > +#if USE_NEARBYINT_BUILTIN > + return __builtin_nearbyint (x); > +#else > + /* Use generic implementation. */ > fenv_t env; > int64_t i0, sx; > int32_t j0; > @@ -67,5 +74,6 @@ __nearbyint (double x) > math_force_eval (t); > libc_fesetenv (&env); > return t; > +#endif /* USE_NEARBYINT_BUILTIN */ > } > libm_alias_double (__nearbyint, nearbyint) Ok. > diff --git a/sysdeps/ieee754/float128/float128_private.h b/sysdeps/ieee754/float128/float128_private.h > index 4e31ef365b..e96986a968 100644 > --- a/sysdeps/ieee754/float128/float128_private.h > +++ b/sysdeps/ieee754/float128/float128_private.h > @@ -139,6 +139,9 @@ > #undef libm_alias_double_ldouble > #define libm_alias_double_ldouble(func) libm_alias_float64_float128 (func) > > +#include > +#undef USE_NEARBYINTL_BUILTIN > +#define USE_NEARBYINTL_BUILTIN USE_NEARBYINTF128_BUILTIN Ok. > > /* IEEE function renames. */ > #define __ieee754_acoshl __ieee754_acoshf128 > @@ -342,6 +345,7 @@ > /* Builtin renames. */ > #define __builtin_copysignl __builtin_copysignf128 > #define __builtin_signbitl __builtin_signbit > +#define __builtin_nearbyintl __builtin_nearbyintf128 > > /* Get the constant suffix from bits/floatn-compat.h. */ > #define L(x) __f128 (x) Ok. > diff --git a/sysdeps/ieee754/flt-32/s_nearbyintf.c b/sysdeps/ieee754/flt-32/s_nearbyintf.c > index acedf34c90..d31bbdd19a 100644 > --- a/sysdeps/ieee754/flt-32/s_nearbyintf.c > +++ b/sysdeps/ieee754/flt-32/s_nearbyintf.c > @@ -21,42 +21,56 @@ > #include > #include > #include > +#include > > +#if ! USE_NEARBYINTF_BUILTIN > static const float > -TWO23[2]={ > - 8.3886080000e+06, /* 0x4b000000 */ > - -8.3886080000e+06, /* 0xcb000000 */ > +TWO23[2] = { > + 8.3886080000e+06, /* 0x4b000000 */ > + -8.3886080000e+06, /* 0xcb000000 */ > }; > +#endif You could just move it inside the !USE_NEARBYINT_BUILTIN within the function. > > float > -__nearbyintf(float x) > +__nearbyintf (float x) > { > - fenv_t env; > - int32_t i0,j0,sx; > - float w,t; > - GET_FLOAT_WORD(i0,x); > - sx = (i0>>31)&1; > - j0 = ((i0>>23)&0xff)-0x7f; > - if(j0<23) { > - if(j0<0) { > - libc_feholdexceptf (&env); > - w = TWO23[sx] + math_opt_barrier (x); > - t = w-TWO23[sx]; > - math_force_eval (t); > - libc_fesetenvf (&env); > - GET_FLOAT_WORD(i0,t); > - SET_FLOAT_WORD(t,(i0&0x7fffffff)|(sx<<31)); > - return t; > - } > - } else { > - if(__builtin_expect(j0==0x80, 0)) return x+x; /* inf or NaN */ > - else return x; /* x is integral */ > +#if USE_NEARBYINTF_BUILTIN > + return __builtin_nearbyintf (x); > +#else > + /* Use generic implementation. */ > + fenv_t env; > + int32_t i0, j0, sx; > + float w, t; > + GET_FLOAT_WORD (i0, x); > + sx = (i0 >> 31) & 1; > + j0 = ((i0 >> 23) & 0xff) - 0x7f; > + if (j0 < 23) > + { > + if (j0 < 0) > + { > + libc_feholdexceptf (&env); > + w = TWO23[sx] + math_opt_barrier (x); > + t = w - TWO23[sx]; > + math_force_eval (t); > + libc_fesetenvf (&env); > + GET_FLOAT_WORD (i0, t); > + SET_FLOAT_WORD (t, (i0 & 0x7fffffff) | (sx << 31)); > + return t; > } > - libc_feholdexceptf (&env); > - w = TWO23[sx] + math_opt_barrier (x); > - t = w-TWO23[sx]; > - math_force_eval (t); > - libc_fesetenvf (&env); > - return t; > + } > + else > + { > + if (__glibc_unlikely (j0 == 0x80)) > + return x + x; /* inf or NaN */ > + else > + return x; /* x is integral */ > + } > + libc_feholdexceptf (&env); > + w = TWO23[sx] + math_opt_barrier (x); > + t = w - TWO23[sx]; > + math_force_eval (t); > + libc_fesetenvf (&env); > + return t; > +#endif /* USE_NEARBYINT_BUILTIN */ > } > libm_alias_float (__nearbyint, nearbyint) Ok, but fix the indentation in a separted patch. > diff --git a/sysdeps/ieee754/ldbl-128/s_nearbyintl.c b/sysdeps/ieee754/ldbl-128/s_nearbyintl.c > index f044cb4334..8f3a7e8e6b 100644 > --- a/sysdeps/ieee754/ldbl-128/s_nearbyintl.c > +++ b/sysdeps/ieee754/ldbl-128/s_nearbyintl.c > @@ -28,42 +28,56 @@ > #include > #include > #include > +#include > > +#if ! USE_NEARBYINTL_BUILTIN > static const _Float128 > -TWO112[2]={ > - L(5.19229685853482762853049632922009600E+33), /* 0x406F000000000000, 0 */ > - L(-5.19229685853482762853049632922009600E+33) /* 0xC06F000000000000, 0 */ > +TWO112[2] = { > + L(5.19229685853482762853049632922009600E+33), /* 0x406F000000000000, 0 */ > + L(-5.19229685853482762853049632922009600E+33) /* 0xC06F000000000000, 0 */ > }; > +#endif You could just move it inside the !USE_NEARBYINT_BUILTIN within the function. > > -_Float128 __nearbyintl(_Float128 x) > +_Float128 __nearbyintl (_Float128 x) > { > - fenv_t env; > - int64_t i0,j0,sx; > - uint64_t i1 __attribute__ ((unused)); > - _Float128 w,t; > - GET_LDOUBLE_WORDS64(i0,i1,x); > - sx = (((uint64_t)i0)>>63); > - j0 = ((i0>>48)&0x7fff)-0x3fff; > - if(j0<112) { > - if(j0<0) { > - feholdexcept (&env); > - w = TWO112[sx] + math_opt_barrier (x); > - t = w-TWO112[sx]; > - math_force_eval (t); > - fesetenv (&env); > - GET_LDOUBLE_MSW64(i0,t); > - SET_LDOUBLE_MSW64(t,(i0&0x7fffffffffffffffLL)|(sx<<63)); > - return t; > - } > - } else { > - if(j0==0x4000) return x+x; /* inf or NaN */ > - else return x; /* x is integral */ > +#if USE_NEARBYINTL_BUILTIN > + return __builtin_nearbyintl (x); > +#else > + /* Use generic implementation. */ > + fenv_t env; > + int64_t i0, j0, sx; > + uint64_t i1 __attribute__ ((unused)); > + _Float128 w, t; > + GET_LDOUBLE_WORDS64 (i0, i1, x); > + sx = (((uint64_t) i0) >> 63); > + j0 = ((i0 >> 48) & 0x7fff) - 0x3fff; > + if (j0 < 112) > + { > + if (j0 < 0) > + { > + feholdexcept (&env); > + w = TWO112[sx] + math_opt_barrier (x); > + t = w - TWO112[sx]; > + math_force_eval (t); > + fesetenv (&env); > + GET_LDOUBLE_MSW64 (i0, t); > + SET_LDOUBLE_MSW64 (t, (i0 & 0x7fffffffffffffffLL) | (sx << 63)); > + return t; > } > - feholdexcept (&env); > - w = TWO112[sx] + math_opt_barrier (x); > - t = w-TWO112[sx]; > - math_force_eval (t); > - fesetenv (&env); > - return t; > + } > + else > + { > + if (j0 == 0x4000) > + return x + x; /* inf or NaN */ > + else > + return x; /* x is integral */ > + } > + feholdexcept (&env); > + w = TWO112[sx] + math_opt_barrier (x); > + t = w - TWO112[sx]; > + math_force_eval (t); > + fesetenv (&env); > + return t; > +#endif /* USE_NEARBYINTL_BUILTIN */ > } > libm_alias_ldouble (__nearbyint, nearbyint) Ok, but fix the indentation in a separated patch. > diff --git a/sysdeps/s390/fpu/math-use-builtins.h b/sysdeps/s390/fpu/math-use-builtins.h > new file mode 100644 > index 0000000000..fd9da8893e > --- /dev/null > +++ b/sysdeps/s390/fpu/math-use-builtins.h > @@ -0,0 +1,45 @@ > +/* Using math gcc builtins instead of generic implementation. s390/s390x version. > + Copyright (C) 2019 Free Software Foundation, Inc. > + This file is part of the GNU C Library. > + > + The GNU C Library is free software; you can redistribute it and/or > + modify it under the terms of the GNU Lesser General Public > + License as published by the Free Software Foundation; either > + version 2.1 of the License, or (at your option) any later version. > + > + The GNU C Library is distributed in the hope that it will be useful, > + but WITHOUT ANY WARRANTY; without even the implied warranty of > + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU > + Lesser General Public License for more details. > + > + You should have received a copy of the GNU Lesser General Public > + License along with the GNU C Library; if not, see > + . */ > + > +#ifndef MATH_USE_BUILTINS_S390_H > +#define MATH_USE_BUILTINS_S390_H 1 > + > +#ifdef HAVE_S390_MIN_Z196_ZARCH_ASM_SUPPORT > + > +# include /* For __GNUC_PREREQ. */ > + > +/* GCC emits the z196 zarch "load fp integer" instructions for these > + builtins if build with at least --march=z196 -mzarch. Otherwise a > + function call to libc is emitted. */ > +# define USE_NEARBYINT_BUILTIN 1 > +# define USE_NEARBYINTF_BUILTIN 1 > +# define USE_NEARBYINTL_BUILTIN 1 > + > +# if __GNUC_PREREQ (8, 1) > +# define USE_NEARBYINTF128_BUILTIN 1 > +# else > +# define USE_NEARBYINTF128_BUILTIN 0 > +# endif > + > +#else > + > +# include_next > + > +#endif > + > +#endif /* math-use-builtins.h */ > Ok, but since the idea is to just override it by the architecture I think it should be more clear to just replicate the expected USE_NEARBYINT* values for !HAVE_S390_MIN_Z196_ZARCH_ASM_SUPPORT.